ROCm/composable_kernel
1
0
Fork 0
mirror of https://github.com/ROCm/composable_kernel.git synced 2026-05-11 17:00:18 +00:00
Code Issues Packages Projects Releases Wiki Activity
407 Commits 739 Branches 38 Tags
7b1ec41e5bd0a644a74f275149b2d625b3f13982
Commit Graph

407 Commits

This Branch
This Branch
All Branches
Author SHA1 Message Date
Chao Liu
7b1ec41e5b refactor 2021-08-06 20:50:01 +00:00
Chao Liu
49c33aaea7 refactor 2021-08-06 19:59:53 +00:00
Chao Liu
54b3e73d17 rename 2021-08-06 18:07:15 +00:00
Chao Liu
f6edda6119 Merge pull request #3 from ROCmSoftwarePlatform/format 
Update to clang-format-10
 2021-07-30 17:16:23 -05:00
Chao Liu
82fae390fb update to clang-format-10 2021-07-30 16:37:00 -05:00
Chao Liu
bd27ed6c38 Merge pull request #2 from asroy/master 
Update readme
 2021-07-28 09:43:56 -05:00
Chao Liu
85a1429301 Update README.md 2021-07-28 09:41:38 -05:00
Chao Liu
56f93c6f33 Update README.md 2021-07-28 09:40:44 -05:00
Chao Liu
594f1dbe96 Merge pull request #1 from ROCmSoftwarePlatform/some_fix_210727 
fix building issue
 2021-07-27 13:19:11 -05:00
Chao Liu
6a1bc5939c fix building 2021-07-27 13:12:43 -05:00
Chao Liu
f63a23acb1 [MIOpen Downstream] Initial MIOpen integration (#52) 
* update online kernel wrapper bundle all descriptors in a tuple

* change __CONSTANT__ to CONSTANT

* rename

* adding tuning

* added IsValidCompileParameter

* reorginze

* adding tunable for fp16 and int8

* fix kernel compile warning and bug fixes

* suppress warning about cast CONSTANT (address space 4) pointer

* fix building issue
 2021-07-27 00:02:27 -05:00
Chao Liu
1264925422 reorganize files to prepare for MIOpen integration (#51) 
* change olc cmake

* adding online compile to fwd-v4r5r2

* update scripts

* remane fwd-v4r5r2 to fwd-v6r1

* clean up
 2021-07-18 00:43:05 -05:00
zjing14
fbdf4332c7 Add xdlops v4r4r4 into online compilation (#48) 
* init for v4r4 xdlops olc

* refactor wrap

* init impl of v4r4 nchw xdlops olc

* tuning

* test perf

* fixed v4r4 nhwc

* tuned v4r4 nhwc

* use gridwise_gemm_xdlops_v2r3

* swap a/b

* add pointer support into offline v2r3

* debugging v4r4r4 transform for olc

* change timer of olc

* refactor v4r4 xdlops nchw olc

* remove transform fun in v4r4 xdlops nhwc olc

Co-authored-by: Chao Liu <chao.liu2@amd.com>
 2021-07-16 23:27:08 -05:00
Chao Liu
0a72e4df94 Change initialization method of tensor for iGEMM (#49) 
* change init method
 2021-07-16 22:55:01 -05:00
Chao Liu
58a8057011 default iterator hack for blockwise copy (#47) 2021-07-16 08:57:15 -05:00
Chao Liu
1c1b56fe61 fix bug: config for ThreadwiseDynamicTensorSliceTransfer_v2 (#46) 2021-07-09 10:13:20 -05:00
Chao Liu
4682d070a6 Create README.md (#45) 
* Create README.md
 2021-07-08 13:32:29 -05:00
Chao Liu
aafb5eb187 Tweak (#44) 
* tweak
 2021-07-08 12:17:43 -05:00
Chao Liu
2f82cfb190 Update default launch bounds (#43) 
* update default launch bounds
 2021-07-08 11:26:57 -05:00
Chao Liu
81c942cd7e Deprecate static kernel (#42) 
* deprecate static kernels
 2021-07-08 10:40:00 -05:00
Chao Liu
b8b2d0a6d1 DL GEMM fp32/fp16/int8 (#41) 
* add threadwise copy the copy a tensor in one copy, added kpack to DL GEMM

* add kpack into fwd v4r5 nchw fp32
 2021-07-04 22:50:29 -05:00
Chao Liu
11ec07e9d1 fix complain about divide by zero (#40) 2021-07-01 16:50:57 -05:00
zjing14
3835318cc3 xdlops_v4r4_fwd fp32/fp16 (#34) 
* create files for xdlops

* working on blockwise_gemm_xdlops

* add KReduction

* add m/n repeats

* add 2x2 pipeline

* added 128x128 wavegemm

* use StaticBuffer of vector_type

* break vector type to blk_size

* add kpack into xldops_gemm and blockwise_gemm

* abroadcast only

* add fp32 mfma instructions

* adding fp16 mfma

* pack half4_t

* rename kperwave to kpack

* add 32x32x8fp16

* add fp16 mfma

* clean code

* clean code

* V4r4 xdlops kpack (#35)

* add kpack with incorrect results

* bug fix for make_dynamic_naive_tensor_descriptor_aligned_v2

* add 1x1 kernel

* add gridwise_gemm_v2 - single_buffer

* enabled dwordx4 for fp16

Co-authored-by: Chao Liu <chao.liu2@amd.com>

* refactor fwd-v4r4-xdlops

* add v4r4-nhwc-xdlop

* improve some perf of nhwc and nchw by tuning parameters, and change scheuduling in gridwise-gemm loop

* tweak scheduling in gridwise gemm

* add v4r3 with a single output copy

* init commit: output with slice win

* adding sliceWin

* add multiple repeats pattern

* starting adding bwd-v4r1-xdlops

* use tuple as SrcBuffer

* adding bwd-data v4r1 nhwc xdlops

* fix bug in make_dynamic_naive_tensor_descriptor_aligned_v2()

* fix bug in host bwd-data conv

* initial implementation of bwd-data v4r1 nhwc xdlops

* add launch bound flags

* enable launch bound

* add m/nrepeat=4

* tweak bwd-data v4r1 nhwc xdlops

* added bwd-data v4r1 nhwc xlops with output A and weight B

* add fwd-v4r4 nhwc xdlops, A input, B weight, C output

Co-authored-by: Chao Liu <chao.liu2@amd.com>
 2021-07-01 14:33:00 -05:00
Qianfeng
1685048a67 Add online compilation for dynamic kernels (#37) 
* Add online-compiling facility

* Synchronize from fwd-v4r5 and implement host interfaces to call conv-fwd v4r4/v4r5 using on-line compiling method

* Tiny adjustment to time reporting

* Use object assignment to replace explicit bytes copying in the first kernel of v4r4/v4r5

* Use single thread to assign descriptor object to device memory

* Adjust to the workload assignment of the two kernels of v4r4 (experimental)

* Revert "Adjust to the workload assignment of the two kernels of v4r4 (experimental)"

This reverts commit eb38461456bb0c82b6c0d32cdd616e181907e20c.

* Update to make constexpr for generating descriptor types in kernel 2 of dynamic conv-fwd v4r4

* Update to dynamic conv-fwd v4r4 online-compiling

* Update to dynamic conv-fwd v4r5 online-compiling (result not accurate)

* Tiny update to driver/CMakeLists.txt

* clang-format

* Tiny comments change

* Add env OLC_DUMP_SAVE_TMP_DIR to support saving of temperary dir

* Fwd v4r5 olc perf (#39)

* added hip-clang flags that fix perf issue of online compilation

* fix bug for olc fwd-v4r5-nchw

* Move constexpr and type reference statements out of the function body in conv-fwd v4r4/v4r5 kernel wrapper

* Remove printing in hip_build_utils.cpp

* Update to root CMakeLists.txt

* Revert "Move constexpr and type reference statements out of the function body in conv-fwd v4r4/v4r5 kernel wrapper"

This reverts commit 3d2c5d8ecdd8298b72d127110500ed5b38d9835c.

Co-authored-by: Chao Liu <chao.liu2@amd.com>
Co-authored-by: Chao Liu <lc.roy86@gmail.com>
Co-authored-by: root <root@dc-smc-18.amd.com>
 2021-06-24 08:34:19 -05:00
Chao Liu
d2315b0dfc pass-by-void-pointer for gridwise_dynamic_gemm_v1r2 (#38) 
* pass-by-void-pointer for gridwise_dynamic_gemm_v1r2

* use pass-by-value by default
 2021-06-19 13:43:45 -05:00
Chao Liu
30072aec37 Restructure gridwise and blockwise GEMM, add tensor contraction and FWD-v4r5 (#36) 
* experimenting magic number division

* overhauling fwd-v4r4 to clearly reflect transformation graph

* added fwd-v4r5

* bug fix for make_dynamic_naive_tensor_descriptor_aligned_v2

* bug fix and added sanity-check in transform_dynamic_tensor_descriptor

* added conv_driver_v2
 2021-06-09 23:53:08 -05:00
Chao Liu
71d6b19d18 reorganize some files (#33) 2021-05-12 14:15:38 -05:00
Chao Liu
78b987fbd6 Use DynamicBuffer instead of raw pointer (#32) 
* Use DynamicBuffer to hold raw pointer (to global and LDS memory)

* add workaround for compiler issue (inefficient ISA) of ds_write for int8x4, int8x8, int8x16
 2021-05-12 13:10:42 -05:00
Chao Liu
01055d95d9 No raw index calculation (#31) 
* Replace most raw index calculation to coordinate transformation
* Overhaul blockwise and threadwise GEMM
* Overhaul driver for gridwies GEMM kernel

Co-authored-by: Jing Zhang <jizhan@amd.com>
 2021-05-11 00:09:25 -05:00
Chao Liu
d075adf126 Use Tuple and vector_type instead of Array for holding tensor data (#30) 
* replacing array with tuple and vector for tensor data
 2021-04-28 13:10:33 -05:00
Chao Liu
e4790c250c Overhaul vector_type and use real vector for int8x4_t instead of aliasing from int32_t (#29) 
* overhaul vector_type, make int8x4_t real vector instead of aliasing from int32_t
 2021-04-12 23:48:43 -05:00
Chao Liu
3bf52e60c5 Initial implementation of magic number division and "Merge" transformation that use it (#28) 
* initial implementation for magic number division and DynamicMerge_v2_magic_division that uses it

* turn off DynamicMerge_v2_magic_division that use magic number division by default
 2021-04-12 21:32:55 -05:00
zjing14
792a20fa5b Hybrid direct + implicit GEMM forward convolution NCHWc v5r1 (#25) 
* Hybrid direct + implicit GEMM forward convolution NCHWc v5r1. Input tensor bypass LDS. Support fp32/fp16/int8
 2021-04-07 16:47:29 -05:00
Chao Liu
d2217f3040 Fix performance issue when passing tensor descriptor from host to kernel by void pointers (#27) 
* use address_space(4) in kernel signature to fix performance issue when passing tensor descriptor from host to kernel by (void) pointers

* remove passing by pointer* option (only use pass by value or void*)
 2021-04-06 17:49:57 -05:00
zjing14
6a5ea49309 bug fix for buffer resource setting (#26) 2021-04-06 16:59:52 -05:00
Chao Liu
fcbb978828 Dynamic tensor descriptor (#24) 
* support dynamic tensor descriptor

* use buffer load OOB feature for padding case

* add navi support

* add int8x4 inference kernel

Co-authored-by: Chao Liu <chao@ixt-rack-81.local.lan>
Co-authored-by: Jing Zhang <jizhan@amd.com>
 2021-03-25 13:51:11 -05:00
Chao Liu
bbcb67d0aa Bwd Data NHWC (#22) 
* fix buffer_store bug
* remove obsolete kernels
* add bwd-data-v5r1-nhwc
 2020-08-06 12:22:11 -05:00
Chao Liu
ac62d13ecd Improve buffer address for out of bound check (#21) 
* Use buffer load built-in OOB check. buffer size is limited to 2GB.
* buffer APIs use combined wave and thread offset
* use uint32_t for addr shift in buffer addressing
 2020-07-29 18:04:09 -05:00
Chao Liu
5c7cec1115 Code clean up (#20) 
* tuning para,

* testing on v100

* add fp16

* remove deprecated tensor descriptor

* sync with miopen

* update build script

Co-authored-by: Jing Zhang <jizhan@amd.com>
 2020-06-23 20:31:27 -05:00
Chao Liu
7d09790a0a MIOpen integration (#15) 
* renaming
 2020-02-18 10:42:18 -06:00
Chao Liu
1a66e35b6f MIopen integration (#13) 
* update for miopen integration: cosmetic refactor
 2020-02-17 09:53:20 -06:00
Chao Liu
3406a1148a Update for recent MIOpen integration (#11) 
* update for MIOpen integration
 2020-01-27 15:29:33 -06:00
Chao Liu
c5da0377fb Added bwd data v3r1 v4r1, tweaking v1 (#10) 
* Added bwd data v3r1: breaking down compute into a series of load balanced GEMM, and launch in a single kernel
* Added bwd data v4r1: like v3r1, but launch GEMMs in multiple kernels
* Tweaked v1r1  and v1r2 (atomic) on AMD GPU
 2020-01-20 10:20:03 -06:00
Chao Liu
e2b4c5b469 update implicit GEMM forward v4r4 to use gridwise gemm (#9) 
* updated fwd v4r4 to use gridwise gemm
* updated gridwise gemm api calls in bwd-data v1r1 and v2r1
 2019-12-05 12:36:36 -06:00
Chao Liu
19a93dac05 fixed faulty padding API calls (#8) 2019-12-03 01:46:44 -06:00
Chao Liu
8f5f64960e backward data (#7) 
* enabled atomic add in tensor copy
* added gridwise GEMM
* added backward data conv using GEMM + atomic
* added backward data conv using GEMM, no atomic
 2019-12-03 01:16:12 -06:00
Chao Liu
31ded4ac4b remove dead file (#6) 2019-11-04 17:13:38 -06:00
Chao Liu
562e1e2767 MIOpen integration: recent bug fixes from MIOpen (#5) 2019-11-04 16:51:12 -06:00
Chao Liu
52c3fe05be Refactor for MIOpen integration (#4) 
Refactor, so can bring multi-index transformation and padding support into MIOpen
 2019-10-11 11:37:31 -05:00
Chao Liu
9aaeacc82b Merge pull request #3 from asroy/clean_up 
enable type conversion in ThreadwiseGenericTensorSliceCopy_v2r1 and BlockwiseGenericTensorSliceCopy_v2
 2019-09-30 15:14:02 -05:00